Analysis
Matthew Loder
7/11/2019
Cleaning scripts:
Load data:
Compare probability ratings
compare_probabilities = function(d) {
d = d %>%
mutate(prob_diff = abs(100*prob_rating - prob))
d_mean_overall = d %>%
group_by(workerid) %>%
summarise(by_worker = mean(prob_diff))
d_mean_by_prob = d %>%
group_by(prob) %>%
summarize(by_prob = mean(prob_diff))
d_mean_by_probworker = d %>%
group_by(workerid, prob) %>%
summarize(by_probworker = mean(prob_diff))
return(d_mean_by_probworker)
}
compare_probabilities(exp_trials1)## # A tibble: 30 x 3
## # Groups: workerid [10]
## workerid prob by_probworker
## <int> <dbl> <dbl>
## 1 0 25 8.6
## 2 0 60 3.40
## 3 0 100 1
## 4 1 25 19.4
## 5 1 60 19
## 6 1 100 54.7
## 7 2 25 16.2
## 8 2 60 12.6
## 9 2 100 2
## 10 3 25 52
## # … with 20 more rowscompare_probabilities(exp_trials2)## # A tibble: 30 x 3
## # Groups: workerid [10]
## workerid prob by_probworker
## <int> <dbl> <dbl>
## 1 0 25 13
## 2 0 60 13.4
## 3 0 100 21.3
## 4 1 25 5
## 5 1 60 6.60
## 6 1 100 7
## 7 2 25 36.8
## 8 2 60 40.8
## 9 2 100 3
## 10 3 25 43
## # … with 20 more rowscompare_probabilities(exp_trials3)## # A tibble: 30 x 3
## # Groups: workerid [10]
## workerid prob by_probworker
## <int> <dbl> <dbl>
## 1 0 60 7.6
## 2 0 90 15.4
## 3 0 100 0
## 4 1 60 16.8
## 5 1 90 47.8
## 6 1 100 74
## 7 2 60 18.4
## 8 2 90 24.3
## 9 2 100 33.7
## 10 3 60 7.6
## # … with 20 more rowscompare_probabilities(exp_trials4)## # A tibble: 30 x 3
## # Groups: workerid [10]
## workerid prob by_probworker
## <int> <dbl> <dbl>
## 1 0 60 6.8
## 2 0 90 7.12
## 3 0 100 0
## 4 1 60 16.2
## 5 1 90 12.1
## 6 1 100 9
## 7 2 60 17.2
## 8 2 90 10.1
## 9 2 100 1.67
## 10 3 60 15.7
## # … with 20 more rowsVisualize mood ratings
mood1 = format_mood(read.csv("../data/pilot3/04_exp_away_cond1-mood_ratings.csv"))
mood1$Answer.condition = "Optimist"
mood2 = format_mood(read.csv("../data/pilot3/04_exp_away_cond2-mood_ratings.csv"))
mood2$Answer.condition = "Confident"
mood3 = format_mood(read.csv("../data/pilot3/04_exp_away_cond3-mood_ratings.csv"))
mood3$Answer.condition = "Pessimist"
mood4 = format_mood(read.csv("../data/pilot3/04_exp_away_cond4-mood_ratings.csv"))
mood4$Answer.condition = "Cautious"
mood2$workerid = mood2$workerid + max(mood1$workerid) + 1
mood3$workerid = mood3$workerid + max(mood2$workerid) + 1
mood4$workerid = mood4$workerid + max(mood3$workerid) + 1
mood_all = rbind(mood1, mood2, mood3, mood4)
mood1_all = mood_all %>%
filter(type == "mood1") %>%
mutate(mood1 = mood_rating) %>%
mutate(mood_rating = NULL) %>%
mutate(type = NULL)
mood2_all = mood_all %>%
filter(type == "mood2") %>%
mutate(mood2 = mood_rating) %>%
mutate(mood_rating = NULL) %>%
mutate(type = NULL)
mood_all = merge(mood1_all, mood2_all)
mood_by_participant = mood_all
mood_by_participant$diff = mood_all$mood2 - mood_all$mood1
moodp1 = ggplot(data = mood_by_participant) +
geom_bar(mapping = aes(x = workerid, y = diff, fill = Answer.condition), stat = "identity")
moodp1
Exclude random responses
exclude_random = function(d) {
d_overall_means = d %>%
group_by(modal, workerid) %>%
summarise(rating_m_overall = mean(rating))
d_indiv_means = d %>%
group_by(modal,percent_window, workerid) %>%
summarise(rating_m = mean(rating))
d_indiv_merged = merge(d_indiv_means, d_overall_means, by=c("workerid", "modal"))
cors = d_indiv_merged %>%
group_by(workerid) %>%
summarise(corr = cor(rating_m, rating_m_overall))
exclude = cors %>%
filter(corr > 0.75) %>%
.$workerid
print(paste("Excluded", length(exclude), "participants based on random responses."))
d = d %>% filter(!(workerid %in% exclude))
}
d1 = exclude_random(d1)## [1] "Excluded 2 participants based on random responses."d2 = exclude_random(d2)## [1] "Excluded 1 participants based on random responses."d3 = exclude_random(d3)## [1] "Excluded 1 participants based on random responses."d4 = exclude_random(d4)## [1] "Excluded 4 participants based on random responses."Aggregated results
## Individual plots
plot(ps1$by_participant)
plot(ps2$by_participant)
plot(ps3$by_participant)
plot(ps4$by_participant)
AUC Computation
So, we’re seeing what we expected for the confident speaker, with the AUC > in the optimistic condition than in the confident condition (indicating explaining away), but we’re not seeing the same (with just one speaker) for the pessimistic/cautious condition, where we would expect flipped results, with the pessimistic condition having a lower AUC than the cautious condition (with adaptation). To sum up, we expect the order to be, from greater to lower difference, “cautious”, “pessimistic”, “optimistic”, and “confident”.
##
## Two Sample t-test
##
## data: aucs.confident$auc_diff and aucs.cautious$auc_diff
## t = -1.2943, df = 13, p-value = 0.2181
## alternative hypothesis: true difference in means is not equal to 0
## 95 percent confidence interval:
## -33.984306 8.519722
## sample estimates:
## mean of x mean of y
## 3.14106 15.87335##
## Two Sample t-test
##
## data: aucs.pessimist$auc_diff and aucs.cautious$auc_diff
## t = -1.1534, df = 13, p-value = 0.2695
## alternative hypothesis: true difference in means is not equal to 0
## 95 percent confidence interval:
## -49.80913 15.13535
## sample estimates:
## mean of x mean of y
## -1.463541 15.873352Power analysis
m.cautious = mean(aucs.cautious$auc_diff)
m.pessimist = mean(aucs.pessimist$auc_diff)
sd.cautious = sd(aucs.cautious$auc_diff)
sd.pessimist = sd(aucs.pessimist$auc_diff)
d = m.cautious - m.pessimist
d = d/sd.pessimist
d## [1] 0.5629496power.t.test(n=60, delta=d, sd=1, sig.level = 0.05, type="two.sample", alternative="two.sided")##
## Two-sample t test power calculation
##
## n = 60
## delta = 0.5629496
## sd = 1
## sig.level = 0.05
## power = 0.863952
## alternative = two.sided
##
## NOTE: n is number in *each* group